Between January and April 2026, some of the biggest names in software development pulled back the curtain on their daily workflows. We’ve rounded up the most actionable insights from their setups below:
Boris Cherny, the engineer who built Claude Code at Anthropic
Boris runs 10–15 sessions at a time. Five in numbered terminal tabs, another 5–10 in a browser, and a few slow-cooker tasks from his phone. Every terminal tab uses a separate git checkout so nothing conflicts. His setup is, by his own description, "surprisingly vanilla."
Two things you can steal from him:
Plan Mode: Boris never lets a session touch a line of code until he’s approved the blueprint. He haggles with the AI on the design until it’s perfect, then flips the switch to auto-accept. From there, the code usually lands perfectly in one shot.
CLAUDE.md: This is a shared markdown file where the team logs every hiccup. Every mistake becomes a permanent rule.
He advises to always give a session a way to verify its own work. If that feedback loop exists, he says it 2–3X's the quality of the final result. → Read his full thread
Addy Osmani, Engineering Lead at Google
"I am the senior dev; the LLM is there to accelerate me, not replace my judgment." Addy's entire approach rests on that single boundary.
Before any code gets written, he builds a full written breakdown of what needs to happen. A reasoning model converts that into a sequence of small, scoped tasks. He works through them one at a time, testing after each before moving forward.
His take on why most engineers are struggling? They’re dumping too much onto the tool at once. He treats his sessions like a high-speed junior developer: capable and quick, but someone who needs a clear, firm hand on the wheel at every step of the way. → Read his full post
Daniel Olshansky, former engineer at Google, Twitter, Waymo, and Pocket Network
Daniel quit writing code by hand entirely 6 months ago and hasn't looked back. "We can finally focus on what actually matters: product and engineering."
He runs a multi-model bullpen through iTerm2: 4 Claude instances, 2 Gemini sessions, and Codex via CLI. He treats them like a team with specific roles:
Codex is the Project Manager (great at mapping out the road).
Gemini is the Architect (strong on hard problems, design docs, and complex bugs).
Claude is the Builder (the workhorse shipping the actual cod).
He's deliberate about routing tasks to the right one. → Read his full post
Mikayla Maki, Software Engineer at Zed
Mikayla brings a much-needed reality check from the front lines of production code. Her rule is simple: Never ask a tool to do something you don't already know how to do yourself.
If you can’t audit the code, you can’t catch the bugs. "LLMs automate typing, not thinking." The craft hasn't changed. The team at Zed now just has more time for it. → Read her full post
Andrej Karpathy, former Director of AI at Tesla and founding member of OpenAI
In just a few weeks at the end of 2025, he went from 80% manual coding to the exact opposite. He calls it "Programming in English," and says it's the biggest change to his workflow in roughly two decades of programming.
He’s not wearing rose-colored glasses, though. He notes that the mistakes have evolved, and they’re now subtle conceptual slips rather than simple typos. These tools are like a hasty junior developer that won't ask for directions. You have to watch them like a hawk, but Karpathy admits he could never go back to the old way. → Read his full post
In Part 2, we’re covering Mitchell Hashimoto, Freek Van der Herten, Alexey Grigorev, and DHH. Stay tuned!
